Scientists teaching robots to more accurately detect dairy disease

Lincoln University computer scientists have been involved in modelling a way to better control and manage a disease which costs New Zealand dairy farmers $280 million a year.

Professor Sandhya Samarasinghe and Professor Don Kulasiri are co-authors of a study*, conducted with data from a commercial robotic dairy farm, which designed and built a computational model to help efficient and accurate detection of mastitis in dairy cattle herds.

The robot milkers use sensors to detect the disease.

The study posited the farm’s performance could be improved significantly by decreasing the number of false positive cases they picked up.

A Deep Neural Network (DNN) was used to build, train and validate a classification model using variable combinations, including variables that have not been studied before.

The model met ISO (International Standards Organization) minimum limits while outperforming other past neural network models and reduced the problem of false positive alerts to just 3% (false negative to 1%), while maintaining an exceptionally good capability of detecting clinical cases within the herd.

Dairy cattle mastitis is one of the most notable and costly diseases in the dairy industry worldwide. It severely affects dairy cattle, resulting in costly treatment and causing a huge decrease in milk produced from sick cows.

Accurate mastitis detection helps cut treatment costs, control the disease, retain milk production levels and maintain milk quality grade.

In addition to cutting financial costs, efficient detection helps protect cows and relieve them of the pain of mastitis.

The total mastitis cost to dairying includes the drop in milk production, low-grade milk quality, cattle treatment cost and other costs.

Research data were collected from a commercial Voluntary Milking System (VMS) dairy farm with 24 DeLaval robots, milking about 1,900 cows in a single 13,000mbarn for one year.

During that period, the total number of recorded milking instances was more than 1.1 million.

Neural networks mastitis detection models presented in previous studies were built using data collected from research dairy farms (conventional or robotic), using small herds (100-400 cows).

The authors said their model being built and tested against data collected from a large-scale commercial dairy farm, made it “more realistic and representational”.

 *Detection of dairy cattle Mastitis: modelling of milking features using deep neural networks (available on request)

Source:  Lincoln University

 

Author: Bob Edlin

Editor of AgScience Magazine and Editor of the AgScience Blog